Summary for the WebSphere Application Server performance tests
After running the WebSphere® Application Server performance tests on our test environment, we compiled a summary of our results and recommendations.
Our test results and recommendations are specific to our environment. Parameters useful in our environment might be useful in other environments, but are dependent on application usage and system configuration. You will need to determine what works best for your environment. For our detailed test results information, see Results for the WebSphere Application Server performance tests.
The following are our summary results:
- WebSphere Application
Server version 6.1.0.0 versus version 6.1.0.11
From a performance perspective, WebSphere Application Server version 6.1.0.11 behaves the same as version 6.1.0.0.
- Red Hat distributions
- We recommend using RHEL 4.5 or higher for WebSphere Application Server workloads. We were not able to obtain RHEL 4.4 data because of performance issues which are fixed in RHEL 4.5
- RHEL 4.5 versus RHEL 5.0
From a performance perspective, the WebSphere Application Server environment on RHEL 4.5 behaves the same as when it is on RHEL 5.0.
- I/O latencies
- Disk I/O
We showed that an application like Trade has a low dependency on disk I/O bandwidth as long as the buffer pools from the database are large enough. Moving the database, which is the only component doing disk I/O, to a RAM disk, which could be considered as the fastest disk I/O device possible, gave only a very slight improvement. Which demonstrates that the disk I/O bandwidth was no bottle neck in our tests.
Note: This was done just for testing purposes. Never use a non persistent disk device for any component of a database! - Network
- We showed in a previous paper that the network connection between the application server and the database has a significant impact on throughput. Therefore, in most cases, we used HiperSockets for our network connections (see http://publib.boulder.ibm.com/infocenter/systems/topic/liaag/WASbaseperformance/publishedwasperf37.htm#PublishedWASPerf-gen36 for details).
- Increasing the number of network buffers on the network interfaces improved throughput by 3%.
- Disk I/O
- Scaling the number of CPUs from one to eight gave us very linear throughput and cost scaling, with a scaling factor close to the ideal case. We were able to saturate the WebSphere Application Server CPUs for each scaling point.